New international framework: the sibling Hugging Face application checkout now
contains framework/, a connected citizen and government web application. From that
checkout run python -m framework.server --demo, then open http://127.0.0.1:8787.
All app launch buttons now link to the workspace, embedded in the GitHub Pages site,
which offers a local preview and the verified Hugging Face application host
https://santanubanerjee-taxdirection.hf.space. The Hugging Face checkout includes a
Docker entrypoint so that the existing Space can serve the same framework after
publication and a successful build. The implementation and
documentation cover annual drafts/locks, 15 country configurations, transactional
batch heatmaps, proposal provenance, government responses and production requirements.
The country registry does not
imply live verification or official endorsement.
The previous public site is preserved in legacy.html. See CONTENT_AUDIT.md for the source-by-source preservation check.
"Revolutionizing Tax from a scalar to a vector."
#TaxDirection lets Indian taxpayers direct a small, government-allowed fraction of their tax (default 5%) toward the specific problems they most want solved. Taxpayers log in, verify their PAN, declare the year's tax, and submit up to 3 "wishes" — each with a percentage weight (the three auto-normalize to 100%). Wishes are aggregated and clustered by problem-domain and geography; each cluster gets a financial weight (the sum of its wishes' directed amounts), and an LLM generates one project proposal per cluster.
It turns a single tax number into a direction — a vector of citizen priorities — giving the government more data, insight, and accountability while still retaining full control over whether (and how much) to apply the rule.
| Part | Repo / Host | What it is |
|---|---|---|
| Public site + dashboard | this repo → GitHub Pages | A static landing/explainer page (index.html) and a live public results dashboard (dashboard.html) that renders the aggregated clusters and their AI proposals. |
| App + AI pipeline | Hugging Face Space | The Docker-hosted international framework and its Python API, with the older Gradio/NLP code retained. |
index.html— landing page / explainer for the concept and the weighting model, with a styled walkthrough and a link to the app.dashboard.html+dashboard.js— the public results dashboard. It reads the aggregatedclusterstable from Supabase using only the public anon key (guarded by Row-Level Security) and renders one card per cluster: problem/location labels, financial weight, wish count, and the AI-generated proposal.config.js— paste yourSUPABASE_URLandSUPABASE_ANON_KEYhere to connect the dashboard. The anon key is safe to commit; the service-role key is never used in this repo. Until configured, the dashboard shows a friendly "Backend not configured" banner.styles.css— shared styling for the site.
normalized_weight_pct = raw_weight_pct * 100 / sum(raw weights of that user's wishes)
directed_amount_inr = declared_tax_inr * (GOVT_ALLOWED_PCT / 100) * (normalized_weight_pct / 100)
GOVT_ALLOWED_PCT (default 5) is an admin setting, never editable by the user. PAN is validated, then stored only as a sha256 hash plus the last 4 characters — the raw PAN is never stored.
- Live dashboard: https://santanban.github.io/TaxDirection/
- App (Hugging Face Space): https://huggingface.co/spaces/SantanuBanerjee/TaxDirection
- Full free-tier setup (Supabase + Groq + HF + GitHub Pages): see
SETUP.mdin the Hugging Face repo.
Tagging public wishes globally and creating weighted investment of individual tax amounts based on the InSociEUP philosophy to life.
Conference, Venue & Date: 10th Management Doctoral Colloquium & VGSOM Research Scholars Day (MDC-VRS 2024); Vinod Gupta School of Management, IIT Kharagpur, India: https://www.linkedin.com/posts/vgsomiitkgp_mdc-vrs-2024-activity-7142869683329810432-Hc_y
Presentation Date: 1st February, 2024
Track: Finance and Accounting
The index.html and styles.css files are for the website: https://santanban.github.io/TaxDirection/
New pipeline with UI can be found here: https://huggingface.co/spaces/SantanuBanerjee/TaxDirection
(For sample datasets on HuggingFace, we used the MyGov portal submissions: https://github.com/SanTanBan/MyGov-Idea-Box-Suggestions-Scraping)
OK Challenge Page: https://www.omdena.com/chapter-challenges/tax-allocation-manager-system
Chapter Title: Aggregating Citizens’ Personalised Recommendations into Tax Directions for automatic Project Proposal development using NLP, enabling Sustainable Hyperlocal Governance and Longevity
Authors: Santanu Banerjee, Tiyasa Mishra, Anandita Das, Atul Kumar Dubey, Krishna Das
Editor: Prof. Dr. Ramesh Chandra Das; Publisher: Routledge (Taylor & Francis)
ISBN: 9781041025276 ; Chapter No.: 15
The current proposal in FOSS 2026 is to develop an upgraded deployable version of the HuggingFace implementation and brainstorm on how to verify users based on PAN/Taxation...
Project Link: https://fossunited.org/hack/fosshack26/p/c6s7kea2an
I attended Zu~Grama (https://zugrama.org/) in January-Feb 2025 and discussed the idea with an attendee Prof. M. V. Rajeev Gowda, a former Member of Parliament for Karnataka in the Rajya Sabha from 2014 to 2020 and a distinguished professor at IIM Bangalore from 2000 to 2014 (https://www.oneindia.com/politicians/prof.-m.-v.-rajeev-gowda-76817.html; https://en.wikipedia.org/wiki/Rajeev_Gowda). He actually was interested, and himself proposed that "maybe not 100% of the amount, but say 5% can be allocated to projects of their choice". Before this I believed that this Tax~Direction could just be a good feedback system to tell the government regarding which decision to make (per year, through the budgets and policies); but actually allowing 5% of every person's tax amount to be allocated to the specific use case which the individual wants the government to solve (maybe a very specific problem) could be a gamechanger. Therefore, it is actually possible in leading democracies to pitch for a similar 5-10% Tax~Direction based Yearly Budgeting (yes, through this feedback every taxpayer will be able to affect each year's financial budget) and (unbiased) AI-enabled holistic policy development.. The government still retains the power to decide whether to implement the 5% or some-per cent rule; so this is just a soft control, but will probably be for the better than the worse due to more data, insights and accountability...